The surprise of the past year is what did not happen: closed labs hold far more training compute yet opened no growing benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → margin over open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition → models, which keep pace using abundant talent and adequate compute.
Closed models still win on qualities benchmarks do not capture — robustness and general usefulness when every user brings a new problem. That is why they dominate direct assistant work while scoring near open models.
The prediction is financial, not technical: Chinese open-weight labs hit funding trouble first, possibly within the year, and their capability trajectories visibly diverge 3 to 9 months after the money does.
The reinforcement-learning era made distribution a capability input. Labs whose users run Claude Code or Codex can train on real task feedback — the first clear technical area where closed models can pull ahead of open weights.
Banning open models above a compute threshold is unenforceable: training a near-frontier model is cheap next to large-scale deployment, so another sovereign trains it, releases it, and it enters your market with less oversight.
Terms in this piece · Glossary
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Why it matters
Concrete predictions on the open-model landscape from one of the field's closest trackers — useful input for model-strategy planning.
Key quotes
“It’s surprising that the top closed models did not show a growing capability margin over open models , based on compute differences for training and research, especially in the second half of 2025 and through today.”
“The open vs. closed model race, as monitored through benchmarks, will largely be a game of economic staying power and fast-following, until the market structure constricts.”
“I expect Chinese open-weight labs to face funding difficulties first, as soon as later this year.”
“Only real AI revenue begets more investment, eventually that’ll be linked to the ability to keep improving models at a rapid rate.”
